kaag 0.1.0

Knowledge and Aptitude Augmented Generation framework

KAAG: Knowledge and Aptitude Augmented Generation

KAAG is a framework for creating adaptive AI agents that engage in dynamic, context-aware conversations. It integrates knowledge retrieval, aptitude modeling, and large language models to provide a flexible system for complex interactions.

Features

• Dynamic Bayesian Networks (DBNs) for modeling conversation states
• Gamified Interaction Model (GIM) for managing interaction context
• Integration with various LLM providers (OpenAI, Anthropic, Ollama)
• Customizable analyzers for interaction state analysis
• Flexible configuration system for conversation stages and metrics
• Simulation capabilities for testing and evaluating AI agents

Installation

pip install kaag

For development:

git clone https://github.com/aroundAI/kaag.git
cd kaag
pip install -e ".[dev]"

Quick Start

from kaag import KAAG, RAG, NoRAG
from kaag.llm.ollama import OllamaLLM
from kaag.knowledge_retriever.text_file import TextFileKnowledgeRetriever
from kaag.utils.config import load_config
from jinja2 import Environment, FileSystemLoader

# Load configuration and initialize components
config = load_config("config.yaml")
llm = OllamaLLM(model="llama2", api_url="http://localhost:11434")
knowledge_retriever = TextFileKnowledgeRetriever("knowledge.txt", top_k=3)

# Load templates
env = Environment(loader=FileSystemLoader("templates"))
kaag_template = env.get_template('kaag.jinja')
rag_template = env.get_template('rag.jinja')
norag_template = env.get_template('norag.jinja')

# Initialize agents
kaag_agent = KAAG(llm, config, kaag_template)
rag_agent = RAG(llm, config, knowledge_retriever, rag_template)
norag_agent = NoRAG(llm, config, norag_template)

# Process user input
user_input = "Hello, I'm interested in your product."
kaag_response = kaag_agent.process_turn(user_input)
rag_response = rag_agent.process_turn(user_input)
norag_response = norag_agent.process_turn(user_input)

print("KAAG response:", kaag_response)
print("RAG response:", rag_response)
print("NoRAG response:", norag_response)

Documentation

For full documentation, visit docs.kaag.io.

Evaluation

To run evaluations:

python -m scripts.evaluation <num_runs>

Results will be saved in the results directory.

Contributing

We welcome contributions! Please see our Contributing Guide for more details.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Citation

If you use KAAG in your research, please cite:

@article{chaudhuri2023kaag,
  title={Knowledge and Aptitude Augmented Generation: Adaptive Multi-Turn Interaction in LLM Systems},
  author={Chaudhuri, Shauryadeep},
  journal={AroundAI},
  year={2023}
}

Contact

For questions and support, please open an issue on the GitHub repository or contact [e-mail].